Yandex: CatBoost

CatBoost is a method of machine learning.

On July 18, 2017 the Yandex company announced creation of a method of machine learning of CatBoost. It is intended for training of models at diverse data.

In a basis the location information of the user, the history of transactions and type of the device comes. The library of machine learning CatBoost is published in open access, everyone can use it.

CatBoost is stated how the successor of a method of machine learning of Matriksnet - it is applied almost in all services of Yandex. As well as Matriksnet, CatBoost uses the mechanism of a gradient busting (engl. boosting is improvement): it is suitable for work with diverse data.

CatBoost considers models of numeric and non-numerical data - types of clouds or types of buildings. Before these data were translated into a language of digits, and it could change their essence, affect the accuracy of work of model. Now they can be used in original form. It helps to show to CatBoost the increased quality of training. It can be applied in different spheres - from bank to production.

Yandex is engaged many years in machine learning, and CatBoost was created by the best specialists in this area. Uploading CatBoost library publicly, we want to make the contribution to development of machine learning. It is necessary to tell that CatBoost is the Russian method of machine learning which became available in open source. We hope that the community of specialists will appreciate it and will help to make even better. Mikhail Bilenko, chief of the department of machine intelligence and researches of Yandex

The method is tested on services of Yandex. Within the experiment it was applied to improvement of search results, ranging of a tape of recommendations of Yandex. A zen and for forecast calculation of weather in Meteum technology. In further CatBoost will work also at other services. It is used by the Yandex Data Factory command — in the solutions for the industry, in particular for optimization of a consumption of raw materials and prediction of defects. The European center of nuclear researches (CERN) implemented CatBoost: the center uses a product for consolidation of the data obtained from different parts of the detector LHCb.

For work with CatBoost it is enough to set of it on computer. The library supports operating systems Linux, Windows and macOS and is available in the Python and R programming languages.

Loading of CatBoost is available on GitHub.

Robotics